data visualization (dsc 530/cis 602-02)
TRANSCRIPT
Data Visualization (DSC 530CIS 602-02)
Data and Tasks
Dr David Koop
D Koop DSC 530 Spring 2017
Programmatic SVG Examplebull Draw a horizontal bar chart
- var a = [6 2 6 10 7 18 0 17 20 6]
bull Steps - Programmatically create SVG - Create individual rectangle for
each item bull Possible solution
- httpcodepeniodakooppenGrdBjE
2D Koop DSC 530 Spring 2017
Databull What is this data
bull Semantics real-world meaning of the data bull Type structural or mathematical interpretation bull Both often require metadata
- Sometimes we can infer some of this information - Line between data and metadata isnrsquot always clear
3D Koop DSC 530 Spring 2017
22
Fieldattribute
item
Items amp Attributes
4D Koop DSC 530 Spring 2017
Items amp Links
5D Koop DSC 530 Spring 2017
[Bostock 2011]
Item
Links
Positions and Grids
6D Koop DSC 530 Spring 2017
Position Grid
Dataset Types
7D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Programmatic SVG Examplebull Draw a horizontal bar chart
- var a = [6 2 6 10 7 18 0 17 20 6]
bull Steps - Programmatically create SVG - Create individual rectangle for
each item bull Possible solution
- httpcodepeniodakooppenGrdBjE
2D Koop DSC 530 Spring 2017
Databull What is this data
bull Semantics real-world meaning of the data bull Type structural or mathematical interpretation bull Both often require metadata
- Sometimes we can infer some of this information - Line between data and metadata isnrsquot always clear
3D Koop DSC 530 Spring 2017
22
Fieldattribute
item
Items amp Attributes
4D Koop DSC 530 Spring 2017
Items amp Links
5D Koop DSC 530 Spring 2017
[Bostock 2011]
Item
Links
Positions and Grids
6D Koop DSC 530 Spring 2017
Position Grid
Dataset Types
7D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Databull What is this data
bull Semantics real-world meaning of the data bull Type structural or mathematical interpretation bull Both often require metadata
- Sometimes we can infer some of this information - Line between data and metadata isnrsquot always clear
3D Koop DSC 530 Spring 2017
22
Fieldattribute
item
Items amp Attributes
4D Koop DSC 530 Spring 2017
Items amp Links
5D Koop DSC 530 Spring 2017
[Bostock 2011]
Item
Links
Positions and Grids
6D Koop DSC 530 Spring 2017
Position Grid
Dataset Types
7D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
22
Fieldattribute
item
Items amp Attributes
4D Koop DSC 530 Spring 2017
Items amp Links
5D Koop DSC 530 Spring 2017
[Bostock 2011]
Item
Links
Positions and Grids
6D Koop DSC 530 Spring 2017
Position Grid
Dataset Types
7D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Items amp Links
5D Koop DSC 530 Spring 2017
[Bostock 2011]
Item
Links
Positions and Grids
6D Koop DSC 530 Spring 2017
Position Grid
Dataset Types
7D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Positions and Grids
6D Koop DSC 530 Spring 2017
Position Grid
Dataset Types
7D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Dataset Types
7D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Dataset Types
8D Koop DSC 530 Spring 2017
Tables
Attributes (columns)
Items (rows)
Cell containing value
Networks
Link
Node (item)
Trees
Fields (Continuous)
Attributes (columns)
Value in cell
Cell
Multidimensional Table
Value in cell
Grid of positions
Geometry (Spatial)
Position
Dataset Types
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Assignment 1bull httpwwwcisumassdedu
~dkoopdsc530assignment1html bull Use HTML CSS SVG and
JavaScript bull Part 3 will take longer bull Due next Friday (Feb 10) bull Start soon
9D Koop DSC 530 Spring 2017
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Sets amp Lists
10D Koop DSC 530 Spring 2017
[Daniels httpexperimentsundercurrentcom]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Attribute Types
11D Koop DSC 530 Spring 2017
Attribute Types
Ordering Direction
Categorical Ordered
Ordinal Quantitative
Sequential Diverging Cyclic
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
231 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
12D Koop DSC 530 Spring 2017
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
241 = Quantitative2 = Nominal3 = Ordinal
quantitative ordinal categorical
Categorial Ordinal and Quantitative
13D Koop DSC 530 Spring 2017
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Sequential and Diverging Databull Sequential homogenous range
from a minimum to a maximum - Examples Land elevations ocean
depths bull Diverging can be deconstructed
into two sequences pointing in opposite directions - Has a zero point (not necessary 0) - Example Map of both land
elevation and ocean depth
14D Koop DSC 530 Spring 2017
[Rogowitz amp Treinish 1998]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
lar structure of spirals allows for an easy detection of cyclesand for the comparison of periodic data sets Furthermorethe continuity of the data is expressed by using a spiral in-stead of a circle
31 Mathematical description and types of spirals
A spiral is easy to describe and understand in polar coor-dinates ie in the form The distinctive feature ofa spiral is that f is a monotone function In this work we as-sume a spiral is described by
Several simple functions f lead to well-known types ofspirals bull Archimedeslsquo spiral has the form It has the spe-
cial property that a ray emanating from the origincrosses two consecutive arcs of the spiral in a constantdistance
bull The Hyperbolic spiral has the form It is theinverse of Archimedeslsquo spiral with respect to the origin
bull More generally spirals of the form are calledArchimedean spirals
bull The logarithmic spiral has the form It has thespecial property that all arcs cut a ray emanating fromthe origin under the same angleFor the visualization of time-dependent data Archimedeslsquo
spiral seems to be the most appropriate In most applicationsdata from different periods are equally important Thisshould be reflected visually in that the distance to other peri-ods is always the same
32 Mapping data to the spiral
In general markers bars and line elements can be usedto visualize time-series data similar to standard point barand line graphs on Spiral Graphs For instance quantitativediscrete data can be presented as bars on the spiral or bymarks with a corresponding distance to the spiral Howeversince the x and y coordinate are needed to achieve the generalform of the spiral their use is limited for the display of datavalues One might consider to map data values to small ab-solute changes in the radius ie
Yet we have found this way of visualizing to be ineffec-tive We conclude that the general shape of the spiral shouldbe untouched and other attributes should be used such asbull colourbull texture including line styles and patterns
Figure 1Two visualizations of sunshine intensity using about the same screen real estate and thesame color coding scheme In the spiral visualization it is much easier to compare days to spotcloudy time periods or to see events like sunrise and sunset
r f ϕ( )=
r f ϕ( ) dfdϕ------ 0 ϕ R+isingt=
r aϕ=
2πar a ϕfrasl=
r aϕk=
r aekϕ=
r aϕ bv ϕ( ) b πamax v ϕ( )( )---------------------------lt+=
Cyclic Data
15D Koop DSC 530 Spring 2017
[Sunlight intensity Weber et al 2001]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Semanticsbull The type of data does not tell us what the data means or how it
should be interpreted bull Tables have keysvalues fields have independentdependent vars
16D Koop DSC 530 Spring 2017
Attribute SemanticsKeys vs Values (Tables) or Independent vs Dependent (Fields)
Flat
Multidimensional
Tabl
es
Fiel
ds
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Data Model vs Conceptual Modelbull Data Model raw data that has a specific data type (eg floats)
- Temperature Example [325 540 -173] (floats)bull Conceptual Model how we think about the data
- Includes semantics reasoning- Temperature Example
bull Quantitative [3250 5400 -1730]bull Ordered [warm hot cold]bull Categorical [not burned burned not burned]
17D Koop DSC 530 Spring 2017
[via A Lex 2015]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
ldquoComputer-based visualization systems provide visual representations of datasets designed to help people carry out tasks more effectivelyrdquo
18D Koop DSC 530 Spring 2017
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Tasksbull Why Understand data but what do I want to do with it bull Levels High (ProduceConsume) Mid (Search) Low (Queries) bull Another key concern Who
- Designer lt-gt User (A spectrum) - Complex lt-gt Easy to Use - General lt-gt Context-Specific - Flexible lt-gt Constrained - Varied Data lt-gt Specific Data
19D Koop DSC 530 Spring 2017
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Tasks
20D Koop DSC 530 Spring 2017
Trends
Actions
Analyze
Search
Query
Why
All Data
Outliers Features
Attributes
One ManyDistribution Dependency Correlation Similarity
Network Data
Spatial DataShape
Topology
Paths
Extremes
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
Identify Compare Summarize
tag
Target known Target unknown
Location knownLocation unknown
Lookup
Locate
Browse
Explore
Targets
Why
How
What
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Actions Analyze
21D Koop DSC 530 Spring 2017
Analyze
ConsumePresent EnjoyDiscover
ProduceAnnotate Record Derive
tag
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Visualization for Consumptionbull Discover new knowledge
- Generate new hypothesis or verify existing one - Designer doesnrsquot know what users need to see - why doesnt dictate how
bull Present known information - Presenter already knows what the data says - Wants to communicate this to an audience - May be static but not limited to that
bull Enjoy - Similar to discover but without concrete goals - May be enjoyed differently than the original purpose
22D Koop DSC 530 Spring 2017
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
ldquo[W]e scientists now understand how important emotion is to everyday life how valuable Sure utility and usability are important but without fun and pleasure joy and excitement and yes anxiety and anger fear and rage our lives would be incompleterdquo mdashD Norman (Emotional Design)
23D Koop DSC 530 Spring 2017
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Measuring User Experience in Visualizationbull Memorability Capability of maintaining and retrieving information
[J Brown et al 1977] bull Engagement Emotional cognitive and behavioral connection that
exists at any point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull Enjoyment Feeling that causes a person to experience pleasure [16] Pleasure is recognized with occurrent happiness and excitement which can be explained in terms of belief desire and thought [W A Davis 1982]
24D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
HIGH QUALITY DESCRIPTION
LOW QUALITY
DESCRIPTION
MEMORABLE
FORGETTABLE
Memorability
25D Koop DSC 530 Spring 2017
[M Borkin et al InfoVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Memorability Maps instead of Networks
26D Koop DSC 530 Spring 2017
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
EUROGRAPHICS 2015 H Carr K-L Ma and G Santucci(Guest Editors)
Volume 34 (2015) Number 3
Map-based Visualizations Increase Recall Accuracy of Data
Bahador Saket1 Carlos Scheidegger1 Stephen G Kobourov1 and Katy Boumlrner2
1Department of Computer Science University of Arizona Tucson AZ USA2Department of Information and Library Science Indiana University Bloomington IN USA
Figure 1 We investigate the memorability of relational data represented with node-link (left-side) and map-based (right-side)visualizations shown are a node-link and a map-based visualization with 200 nodes and 500 links from the LastFM dataset
AbstractWe investigate the memorability of data represented in two different visualization designs In contrast to recentstudies that examine which types of visual information make visualizations memorable we examine the effect ofdifferent visualizations on time and accuracy of recall of the displayed data minutes and days after interactionwith the visualizations In particular we describe the results of an evaluation comparing the memorability of twodifferent visualizations of the same relational data node-link diagrams and map-based visualization We findsignificant differences in the accuracy of the tasks performed and these differences persist days after the originalexposure to the visualizations Specifically participants in the study recalled the data better when exposed tomap-based visualizations as opposed to node-link diagrams We discuss the scope of the study and its limitationspossible implications and future directions
1 Introduction
Researchers have long recognized that the visual display ofinformation can be more effective than tables and numericsummaries [Ans73] We also know that different visual de-signs offer significantly different reading precision [CM84]In contrast we do not understand nearly as well the mem-orability of the data that underlies the visualization Is thedesign of a visualization responsible for how well users willremember its content
In this paper we present evidence that different visualdesigns can impact the recall accuracy of the data being vi-sualized Several recent studies have tested the memorabilityof different types of visualizations [BMG10 BARM12MPWG12 VMTW12 IXTO11 BVB13] These seminal
studies focused on which types of visual information arememorable [BVB13] To the best of our knowledge nostudy has yet been performed to assess long-term memorabil-ity of the underlying data represented in these visualizationsIn this paper we focus on two alternative visualizations forrelational data Specifically we compare node-link visualiza-tions to map-based visualizations
Node-link visualizations date back to 1735 and are a stan-dard way of depicting relational datasets In node-link dia-grams entities are depicted as points (typically dots or circles)in low-dimensional space and two related entities are con-nected with a curve (typically a straight-line segment) Clustermembership is typically indicated by filling each circle witha color that is unique for each cluster
c 2015 The Author(s)Computer Graphics Forum c 2015 The Eurographics Association and JohnWiley amp Sons Ltd Published by John Wiley amp Sons Ltd
[B Saket et al EuroVis 2015]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
ISOTYPE Visualizationsbull Study [Haroz et al 2015]
- Want quick understanding and ease of remembering
- Does ISOTYPE help bull Results
- Stacked icons allow both length and quantity encoding
- Icons are more memorable - Images that arent used to show
data are distracting
27D Koop DSC 530 Spring 2017
[Image by O and M Neurath Study by S Haroz et al 2015]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Memorabilitybull Capability of maintaining and retrieving information
[J Brown et al 1977] bull How to measure
- test users bull How long
- short-term intermediate or long-term bull What types of visualizations
- barlinepie networks graphs etc
28D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Engagementbull Emotional cognitive and behavioral connection that exists at any
point in time and possibly over time between a user and a resource [S Attfield et al 2011]
bull How to measure total time spent looking at a chart
29D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Measuring Engagement
30D Koop DSC 530 Spring 2017
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Grid is blurred click for detail
[S Haroz et al 2015]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
We ran 50 subjects on Amazon Mechanical Turk in 200 trials (5 chart types times 2 questions times 20 repetitions) blocked by chart type Each subject was paid 8 US Dollars for the 30-minute study and all participants were from the USA Exp 4 Results All subjects showed over 92 accuracy allowing incorrect responses to be dropped from analysis without substantially affecting statistical power We also collapsed across the lsquoMorersquo vs lsquoFewerrsquo condition to yield approximately 40 trials per chart type per subject As with the previous experiments we analyzed the results within-subject to determine the per-formance relative to that of the simple bar charts
We found a main effect of graph type on response time (F[4 49]=20 p lt 005 ηp2=002) A Tukey HSD-corrected comparison of all the graph types found that only the super-fluous condition was significantly different from the stand-ard bar graph (p lt 005) as can be seen in Fig 13
This result combined with the results of experiment 1 show that superfluous images hurt both memorability and speed of usability of charts EXPERIMENT 5 INITIAL ENGAGEMENT Although speed can be an important benchmark the aim of some visualizations is to make people pause and look ndash as is often the case in news articles Designers often rely on pic-tographs because they are thought to draw the attention of a reader When perusing through a collection of articles an en-ticing visualization may increase the likelihood that an article will be inspected more closely Will an ISOTYPE visualiza-tion be better at capturing attention than a simple bar chart We ran an experiment that simulated how visualizations are commonly encountered in a peripheral glimpse as thumb-nails among a collection of text and other visualizations com-peting for interest Exp 5 Methods Subjects were presented with a 3x3 grid of items (Fig 14) Each item included a short title above a small slightly blurred thumbnail The thumbnail was either a set of sen-tences about the topic from Wikipedia or a chart related to the topic The subjects were given two minutes to look through the thumbnails They could click whichever item in-terested them to view the information in full screen without pixilation or blur Clicking again returned them to the grid where they could repeat the process No limit was placed on the number or duration of views for each item However af-ter the trialrsquos time had finished everything was removed from the screen They were then presented with a button to begin the next trial We selected 36 topics from the previous experimentsrsquo cate-gories and constructed text a bar chart and a stacked picto-graph chart for each Throughout the experiment each sub-ject encountered each topic exactly once (9 items times 4 trials) A trial included 3 bar charts 3 stacked pictograph charts and 3 pieces of text We tracked the start time and duration of each view
10 subjects (4 women) participated in this experiment Be-cause it was implemented as a Windows desktop application it was run in the lab All subjects were undergraduates and were paid 5 US dollars for the 15 minute duration Exp 5 Results We binned the first minute of viewing into one-second inter-vals and found the portion of subjects viewing each type of item Fig 15 shows a linear fit of these results collapsed across trial For the first few seconds most are at the selec-tion grid However the ISOTYPE visualization takes a quick
(A)
(B)
Fig 14 (A) An example of the selection grid for experiment 5 The title is readable but the details of the content are unrecog-nizable beyond the type of information (B) An example text dis-play that can also been seen in the middle right of the selection grid in (A)
Fig 15 ISOTYPE charts are best at initially engaging subjects to inspect information more closely
Measuring Engagement
31D Koop DSC 530 Spring 2017
[S Haroz et al 2015]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Enjoyment Name Voyager
32D Koop DSC 530 Spring 2017
[Wattenberg 2005 wwwbabynamewizardcom]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Measuring Enjoymentbull Difference from engagement (eg may be for a job) bull Self-reporting (eg comparison between different charts bull Measure why someone enjoys a visualization
- Challenge - Focus - Clarity - Feedback - Control - Immersion
33D Koop DSC 530 Spring 2017
[B Saket et al BELIV 2016]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
ldquoVisualizations donrsquot need to be designed for memorability ndash they need to be designed for comprehension For most visualizations the comprehension that they provide need only last until the decision that it informs is made Usually that is only a matter of secondsrdquo mdash S Few
34D Koop DSC 530 Spring 2017
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Reactionbull B Jones (paraphrased) People make decisions using visualizations
but this isnt instantaneous like robots or algorithms they often chew on a decision for a while
bull R Kosara there are cases where people benefit from remembering a visualization (eg health-related visualization)
bull Are there tradeoffs between the characteristics
35D Koop DSC 530 Spring 2017
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Visualization for Productionbull Generate new material bull Annotate
- Add more to a visualization - Usually associated with text but can be graphical
bull Record - Persist visualizations for historical record - Provenance (graphical histories) how did I get here
bull Derive (Transform) - Create new data - Create derived attributes (eg mathematical operations
aggregation)
36D Koop DSC 530 Spring 2017
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
MTA Fare Data Exploration
37D Koop DSC 530 Spring 2017
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Actions Search
bull What does a user know bull Lookup check bearings bull Locate find on a map bull Browse whatrsquos nearby bull Explore where to go (patterns)
38D Koop DSC 530 Spring 2017
Search
Target known Target unknown
Location known
Location unknown
Lookup
Locate
Browse
Explore
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Query
bull Number of targets One Some (Often 2) or All bull Identify characteristics or references bull Compare similarities and differences bull Summarize overview of everything
39D Koop DSC 530 Spring 2017
Query
Identify Compare Summarize
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Targets
40D Koop DSC 530 Spring 2017
Trends
ALL DATA
Outliers Features
ATTRIBUTES
One ManyDistribution Dependency Correlation Similarity
Extremes
NETWORK DATA
SPATIAL DATA
Shape
Topology
Paths
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Analysis Example Different ldquoIdiomsrdquo
41D Koop DSC 530 Spring 2017
[SpaceTree Grosjean et al] [TreeJuxtaposer Munzner et al]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
ldquoIdiomrdquo Comparison
42D Koop DSC 530 Spring 2017
[SpaceTree Supporting Exploration in Large Node Link Tree Design Evolution and Empirical Evaluation Grosjean Plaisant and Bederson Proc InfoVis 2002 p 57ndash64]
SpaceTree
[TreeJuxtaposer Scalable Tree Comparison Using Focus+Context With Guaranteed Visibility ACM Trans on Graphics (Proc SIGGRAPH) 22453ndash 462 2003]
TreeJuxtaposer
Present Locate Identify
Path between two nodes
Actions
Targets
SpaceTree
TreeJuxtaposer
Encode Navigate Select Filter AggregateTree
Arrange
Why What How
Encode Navigate Select
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]
Analysis Example Derivationbull Strahler number
ndash centrality metric for treesnetworks ndash derived quantitative attribute ndash draw top 5K of 500K for good skeleton
43D Koop DSC 530 Spring 2017
[Using Strahler numbers for real time visual exploration of huge graphs Auber Proc Intl Conf Computer Vision and Graphics pp 56ndash69 2002]
Task 1
58
54
64
84
24
74
6484
84
94
74
OutQuantitative attribute on nodes
58
54
64
84
24
74
6484
84
94
74
InQuantitative attribute on nodes
Task 2
Derive
WhyWhat
In Tree ReduceSummarize
HowWhyWhat
In Quantitative attribute on nodes TopologyIn Tree
Filter
InTree
OutFiltered TreeRemoved unimportant parts
InTree +
Out Quantitative attribute on nodes Out Filtered Tree
[Munzner (ill Maguire) 2014]